Content uploaded by Thomas Buck
Author content
All content in this area was uploaded by Thomas Buck on Jan 13, 2016
Content may be subject to copyright.
ORIGINAL PAPER
Automated flow quantification in valvular heart disease
based on backscattered Doppler power analysis:
implementation on matrix-array ultrasound imaging
systems
Thomas Buck ÆShawn M. Hwang ÆBjo
¨rn Plicht ÆRonald A. Mucci Æ
Peter Hunold ÆRaimund Erbel ÆRobert A. Levine
Received: 11 October 2007 / Accepted: 18 February 2008
ÓSpringer Science+Business Media B.V. 2008
Abstract Objective Cardiac ultrasound imaging
systems are limited in the noninvasive quantification
of valvular regurgitation due to indirect measure-
ments and inaccurate hemodynamic assumptions. We
recently demonstrated that the principle of integration
of backscattered acoustic Doppler power times
velocity can be used for flow quantification in
valvular regurgitation directly at the vena contracta
of a regurgitant flow jet. We now aimed to accom-
plish implementation of automated Doppler power
flow analysis software on a standard cardiac ultra-
sound system utilizing novel matrix-array transducer
technology with detailed description of system
requirements, components and software contributing
to the system. Methods This system based on a
3.5 MHz, matrix-array cardiac ultrasound scanner
(Sonos 5500, Philips Medical Systems) was validated
by means of comprehensive experimental signal
generator trials, in vitro flow phantom trials and in
vivo testing in 48 patients with mitral regurgitation of
different severity and etiology using magnetic reso-
nance imaging (MRI) for reference. Results All
measurements displayed good correlation to the
reference values, indicating successful implementa-
tion of automated Doppler power flow analysis on a
matrix-array ultrasound imaging system. Systematic
underestimation of effective regurgitant orifice areas
[0.65 cm
2
and volumes [40 ml was found due to
currently limited Doppler beam width that could be
readily overcome by the use of new generation 2D
matrix-array technology. Conclusion Automated flow
quantification in valvular heart disease based on
backscattered Doppler power can be fully imple-
mented on board a routinely used matrix-array
ultrasound imaging systems. Such automated Doppler
power flow analysis of valvular regurgitant flow
directly, noninvasively, and user independent over-
comes the practical limitations of current techniques.
Keywords Echocardiography
Flow quantification Matrix-array transducer
Magnetic resonance imaging Mitral regurgitation
Spectral Doppler power
T. Buck (&)B. Plicht R. Erbel
West German Heart Center Essen, Department of
Cardiology, University of Duisburg-Essen,
Hufelandstrasse 55, 45122 Essen, Germany
e-mail: thomas.buck@uk-essen.de
S. M. Hwang
Department of Electrical Engineering and Computer
Science, Massachusetts Institute of Technology, Boston,
MA, USA
R. A. Mucci R. A. Levine
Cardiac Ultrasound Laboratory, Massachusetts General
Hospital, Harvard Medical School, Boston, MA, USA
P. Hunold
Institute of Diagnostic and Interventional Radiology,
University of Duisburg-Essen, Essen, Germany
123
Int J Cardiovasc Imaging
DOI 10.1007/s10554-008-9302-8
Introduction
Today, ultrasound has become the most frequently
used diagnostic imaging technique for cardiovascular
disease. Through the use of the Doppler concept,
ultrasound can also be used to obtain information
pertaining to blood flow within the heart for diag-
nostic purposes [1]. Mitral regurgitation (MR) is the
most common valvular heart disease in western
industrial countries [2,3] with disease symptoms
attributed to the volume overload from regurgitant
flow volume within each heart beat, causing and
aggravating heart failure [4]. For optimal timing of
surgical therapy, therefore, assessment of the severity
of MR requires accurate and feasible quantification of
regurgitant flow and effective regurgitant orifice area
(EROA) [5–7]. Existing ultrasound techniques based
on Doppler methods, however, have been unsatisfac-
tory due to invalid hemodynamic assumptions and
indirect estimations, mainly because they cannot
measure regurgitant flow directly at the lesion, which
requires the product of flow velocity and EROA
[8–13]. Building upon the basic research done by
Hottinger et al. [14], we recently showed that since
backscattered Doppler power reflects number of
scatterers, the integral of Doppler power times
velocity (PVI) at the vena contracta (VC) of a
regurgitant jet [15] is proportional to volume flow
rate [16,17]. That analysis, however, was limited by
a beam area which was too narrow as a result of
standard phased-array transducer design and the need
for time-consuming off-line data analysis and manual
tracing of the Doppler spectrum. Therefore we aimed
implementation of the PVI method for automated
flow quantification on a cardiac ultrasound system
utilizing novel matrix-array transducer technology.
Implementation of automated PVI analysis was
considered to exist of two principal components: (1)
special matrix-array transducer technology enabling
electronic control of Doppler beam width, and (2)
automated Doppler power flow analysis software.
Recent experimental validation as well as initial
clinical application demonstrated a high degree of
accuracy between flow measurements and reference
values [18]. The aim of the present work is to provide a
detailed description of the technical requirements and
new concepts underlying practical implementation of
automated PVI analysis, as to be particularly important
for future adaptation to new generation 2D matrix-array
transducer systems that are increasingly used in clinical
echocardiography for real-time 3D imaging [19].
Basic principles and methods
Basic principles of PVI analysis
Quantification of regurgitant flow requires the product
of flow velocity and area of flow [15]. Generally
speaking, the incremental flow d_
Qx;y;tðÞpassing
through the differential area dA
~x;yðÞat time tis given
by v
~x;y;tðÞdA
~x;yðÞ:Thus, the total flow rate through
a planar CSA is given by the integral expression:
_
QvtðÞ¼Z
Al
v
~x;y;tðÞdA
~x;yðÞ ð1Þ
where A
l
denotes the CSA, through which blood
flows with directional velocity function v
~x;y;tðÞ
across the area of flow at time t.
However, existing spectral Doppler analysis tools
only measure velocity, not the area of flow. Standard
spectral Doppler display, in fact, is a three-dimensional
graph, where the x-axis corresponds to time, the y-axis
corresponds to frequency (and also velocity through
the Doppler equation), and each pixel intensity value
corresponds to the backscattered power at a particular
frequency with signal intensity or backscattered power
known to be proportional to the area of flow [1]. To
overcome this limitation of spectral Doppler analysis,
we recently demonstrated a novel approach based on
the Doppler Power principle [14] using backscattered
acoustic power from the spectral Doppler signal to
provide the area information we need [16]: Since
backscattered power (P) is linearly proportional to the
number of scatterers for a given hematocrit [20,21],
backscattered power in the Doppler spectrum of flow
through a thin disk-like sample volume of fixed height
is linearly proportional to the sonified blood volume of
moving scatterers and, therefore, will be linearly
proportional to the CSA of flow as long as the area is
encompassed by the beam [14,22], or:
CSAflow /Power ¼Z
vel
PvðÞdv ð2Þ
where P(v) is the spectral Doppler power associated
with a Doppler frequency. Backscattered power from
non-moving or stagnant blood within the sample
Int J Cardiovasc Imaging
123
volume but outside the flow CSA is eliminated by
high-pass filtering and does not contribute to the
power in the Doppler spectrum from rapidly moving
blood [16].
Because this principle holds only for laminar-like
flow and not for turbulent jets, where for a given flow
rate, backscattered power is nonlinearly increased by
turbulent eddies [20,21] and fluid entrainment into
the jet [15], we applied the Doppler power principle
at the proximal VC (Fig. 1) where flow is laminar
prior to entrainment [16,17]. At the VC of a
regurgitant jet, identified by a narrow, high-velocity
spectrum corresponding to laminar flow (Fig. 2), total
backscattered power integrated over the velocity
spectrum or the power integral (PI) is linearly
proportional to the CSA of the VC or
CSAVC /Z
vel
PvðÞdv ð3Þ
and power times velocity, integrated over the VC
velocity spectrum or the power–velocity integral
(PVI), is proportional to the total instantaneous flow
passing through the VC, denoted _
QvtðÞ[16]or
_
QvtðÞ/Z
vel
vP vðÞdv ð4Þ
In order to obtain clinically useful absolute flow
values, power measurements in each individual can
be calibrated by interrogating a patient with a narrow
reference ultrasound beam of known CSA that fits
entirely within a region of laminar flow at the VC,
according to the previously described attenuation
compensation method [14]. The obtained backscat-
tered power estimate from this narrow beam is
proportional to the known CSA of flow, thereby
providing the appropriate ratio (i.e., proportionality)
between power and CSA for a given individual. This
same ratio is then applied to the power measured by
the broader measurement beam in order to determine
the CSA of laminar flow within the measurement
beam, provided the measurement beam encompasses
the entire CSA of flow, [16] or:
Kcal ¼CSAref
Pref ¼CSAflow
Pmeas ð5Þ
where K
cal
is the calibration coefficient, P
ref
is the
power measurement resulting from the narrow refer-
ence beam; P
meas
is the power measured by the broad
measurement beam, and CSA
flow
is the unknown
CSA of flow within the broad measurement beam
[14,23]. With the two sample volumes of the two
beams placed at the same depth as the pathologic
flow of interest, hematocrit, backscattering coefficient
and attenuation are the same, so that any changes in
backscattered power relate to changes in the CSA of
flow (CSA
flow
) within the beam.
The calibration technique was then applied as
follows to obtain total instantaneous flow passing
through the VC. We first note that flow rate,
Fig. 1 Left, Ultrasound
image of a patient with
mitral valve insufficiency,
mitral regurgitant flow from
the left ventricle (LV)
through the mitral valve
(MV) back into the left
atrium (LA) visualized by
color Doppler imaging. The
dotted scan line indicates
the position of the high PRF
Doppler beam and the
sample volume. Right,
Schematic of the broad
Doppler measurement beam
and sample volume fully
sonifying the regurgitant
flow at the VC
Int J Cardiovasc Imaging
123
_
Q¼Z
vel
vCSAflow vðÞdv ð6Þ
According to Eq. 5,
CSAflow ¼Pmeas
CSAref
Pref
ð7Þ
Thus,
_
Q¼Z
vel
vPmeas vðÞ CSAref
Pref
dv ð8Þ
_
Q¼CSAref
Pref
Z
vel
vPmeas vðÞdv ð9Þ
_
Q¼Kcal Z
vel
vPmeas vðÞdv ð10Þ
The desired total flow volume (i.e., RSV) could
then be obtained from the estimates of instantaneous
flow by integrating over the time interval of interest:
RSV ¼Qv¼Z
T
_
QvtðÞdt ð11Þ
Practical application of the PVI method
expounded above so far has been limited because of
complex offline Doppler signal processing and anal-
ysis and limited beam broadening: Doppler video
display intensities, non-linearly compressed, had
to be reconverted on a PC workstation to their
original uncompressed acoustic amplitudes and
power (amplitude squared) on the basis of the
acquisition compress and reject settings [16]. Using
MATLAB software (Mathworks, Natick, MA, USA)
power and power 9velocity then had to be inte-
grated over all velocities in the manually traced
narrow VC velocity spectrum at each time point.
Furthermore, standard phased-array transducer design
only allowed manual reduction of the transducer
aperture with a mask (Kludgey modification) in order
to create a broader measurement beam of 5.8 mm
width at 10 cm on the basis of the half-maximum-
power beamwidth; this beamwidth yet too small to
sufficiently encompass large or slit-like EROAs [17].
As a result of manual aperture reduction, measure-
ments of the narrow and the broad beam had to be
performed on different Doppler signals, causing 2
disadvantages: (1) As a result of the difference in
aperture size, transmit power and receive sensitivity
of the narrow and broad beam differed from each
other, requiring additional application of a correction
factor (CF) to the calibration coefficient, and (2)
Fig. 2 Automatic velocity
border detection algorithm
at work in the calibration
stage. Left, Narrow-
spectrum Doppler signal
from pulsatile in vitro flow.
Right, Automatic velocity
border detection with
maximum and minimum
velocity borders and
measurement bars—
S(ystole) and D(iastole)—
specifying a cardiac cycle
of interest. Above left,
Window displaying
instantaneous power against
time as well as the mean
power and mean pixel
intensity within the
specified cardiac cycle
Int J Cardiovasc Imaging
123
calibration conditions became potentially incorrect
due to variations of attenuation and flow conditions
over time.
Implementation of Doppler power analysis on
matrix ultrasound imaging systems
To overcome these limitations we developed a
system for automated flow quantification including
(1) entire software implementation of the PVI
algorithms described above on a standard Sonos
5500 ultrasound system, and (2) a special matrix-
array transducer enabling electronic aperture control.
(1) Implementation of PVI software
Implementation of the PVI algorithms included the
following components: (a) automated detection of the
narrow VC Doppler spectrum, (b) availability of
unmodified acoustic amplitude values, (c) implemen-
tation of calibration stage (computation of calibration
coefficient), (d) implementation of the measurement
stage, and (e) practical user interface.
(a) Automated velocity border detection:As
described above all the necessary integrations must
be performed over the narrow velocity spectrum from
flow passing through a thin sample volume placed in
the VC of the regurgitant jet (Fig. 1). Thus, to define
the integration bounds, an algorithm which automat-
ically finds and displays a maximum and a minimum
velocity border around the spectral Doppler wave-
form was incorporated (Fig. 2). Once the integration
bounds were established, just the relevant portions of
the spectral Doppler data could be extracted from
image memory. For those cases where the automatic
border algorithm failed detecting the borders an
option for the user to draw borders manually was also
included.
(b) Recovery of unmodified acoustic amplitude
values: As another intricacy, in the Sonos 5500 used,
like in most other available systems, unmodified
acoustic amplitude values were not directly available
within image memory where the spectral Doppler
data is extracted. That is because the acoustic
amplitude values passed through the 1st stage output
mapping process and then through the post-process-
ing curves before being stored in image memory; the
1st stage output mapping referring to the process of
mapping the fast Fourier transform (FFT) output
values nonlinearly to a range of 0–255 and the
postprocessing, referring to nonlinear video compres-
sion, to utilize effectively the dynamic range of the
display device. Thus, in a first step, to recover
unmodified amplitude values we used a mechanism to
first bypass the nonlinear post-processing curves and
then invert the 1st stage output mapping (Fig. 3).
Only after reworking almost every module along the
signal processing path causing fundamental modifi-
cation of the preexisting system software unmodified
power estimates could be obtained and the required
integrations be performed.
(c) Implementation of the PVI calibration stage:
Within the calibration stage, integration of power
from a narrow reference beam was needed to
calculate the calibration coefficient K
cal
[16]. The
simplest calibration method, in terms of implemen-
tation, would have involved a separate calibration
step, requiring the transmission of a narrow reference
beam, that is, the system would first generate the
narrow reference transmit and receive beams and
then form the appropriate ratio between power and
CSA (see Eq. 5). The disadvantage with this approach
would have been that the transmit power and receive
sensitivity of the narrow and broad beams would
differ from each other [16]. Thus, a correction factor
(CF) would have been required to correct the power
from the measurement beam for the decrease in
transmit powers and receive sensitivities resulting
from aperture reduction. The CF was accounted for in
the ratio given in Eq. 5as such:
Pref
CSAref ¼CF Pmeas
CSAflow ð12Þ
And hence the calibration coefficient, K
cal
, became
the following:
Kcal ¼CF CSAref
Pref ð13Þ
Besides the basic calibration process, software
implementation had to account for system artifacts on
both the receive and the transmit portion of the signal
path such as that changing the velocity scale setting
changed the pulse repetition frequency (PRF), which
led to a change in transmit voltage and energy
delivered to the sample volume, which reciprocally
affected the level of received backscattered signal and,
Int J Cardiovasc Imaging
123
therefore, required correction. As another intricacy,
changing the gain setting between the calibration and
measurement stage in order to optimize Doppler signal
display intensity directly changed the magnitude of the
energy received and thus change of backscattered
power would represent not a change of orifice CSA,
Fig. 3 Complete software control flowchart, with the high-
level software architecture on the left and detailed views of the
main calibration and measurement branches on the right. P(v),
spectral Doppler associated with Doppler frequency/velocity;
P
ref
, power measured by narrow reference beam; K
cal
,
calibration coefficient; CF, correction factor; CSA
ref
, known
cross-sectional area of narrow reference beam; P
avg
=average
of stored P
ref
;P
meas
, power measured by broad measurement
beam; Qv
tðÞ;total instantaneous flow rate; QvðÞ;total flow
volume
Int J Cardiovasc Imaging
123
but rather, merely a change in system settings.
Because undoing the effects of the receive gain or
normalizing to a constant value from a software
standpoint would have been immensely difficult, the
established compromise involved allowing modifica-
tion to the receive gain initially, but locking out
control once the transition from calibration to mea-
surement had been made.
(d) Implementation of the PVI measurement stage:
Up to the processing step where amplitude values are
squared at each frequency bin to get backscattered
power (breakpoint b in Fig. 3), the implementation of
the measurement stage is identical to that of the
calibration stage. After detection of the VC Dop-
pler spectrum, the implementation deviates from the
calibration stage in that the integral of power 9veloc-
ity is being calculated now within the region bounded
by the upper and lower velocity borders. Furthermore,
hence the estimates of flow rate produced by the PVI
are integrated over a time period delineated by the
S(ystole)- and D(iastole)-bars on the spectral Doppler
display (Fig. 2) to compute a measure of total RSV,
the extra dimension of time needed to be factored into
the software calculations as well. Since the metric
generated by each power–velocity integral represents
a differential flow rate at a discrete instant of time,
each flow rate estimate must be multiplied by an
appropriate incremental unit of time before summing.
In effect, the software package is simply computing a
Riemann sum to approximate the continuous integral
written in Eq. 11, and the approximation is given
below in Eq. 14.
RSV ¼Qv¼Z
T
_
QvtðÞdt X
T
_
Qvn½Dtð14Þ
Once all the calculations had been completed,
the software opened a window graphing instanta-
neous flow rate against time, as well as a text
window displaying the total RSV and average flow
CSA computed from the selected cardiac cycle
(Fig. 2).
(e) User interface of automated PVI software:
Taking into account the complexity of the software
that was required for implementation of the PVI
method we aimed to create a user interface making
the clinical application as simple as possible. Thus,
for ease of programming and functional purposes, the
user interface was integrated onto the touch-screen
panel of the ultrasound imaging system. Use of the
software through the user interface to calibrate and to
produce final absolute measures of RSV and flow
CSA is summarized in the following steps:
(1) The user first locates the VC of the MR jet with
the help of color Doppler in the 2D sector scan
mode (Fig. 1). Then the ultrasound system will
be switched to the high-PRF Doppler mode,
1
and the sample gate of the narrow Doppler beam
will be placed at the VC of the regurgitant flow
jet.
(2) After acquisition of a characteristic VC Doppler
spectrum, the ultrasound system will be frozen,
and the main calibration branch of the software
(Fig. 3) will be applied to the spectral Doppler
waveforms including delineation of the laminar
flow portion by either automatic or manual
means.
(3) After selecting a cardiac cycle of interest via the
Doppler measurement bars (Fig. 2), the user
will press the ‘Analyze Cycle’ key. Upon the
keypress, the software will bypass the post-
processing curves and then extract from image
memory the pixel intensity values within the
boundaries set by the upper and lower velocity
borders and the Doppler measurement bars.
After this, reconstructed original amplitude
magnitudes are squared and summed to obtain
an estimate of backscattered power. The power
mean value along with the mean pixel intensity
are then displayed onscreen to the user and
stored by pressing the ‘Store Cycle’ key.
(4) Consequent to the ‘Calculate Coeff.’ keypress,
the calibration coefficient K
cal
will be com-
puted. As secondary effects of the ‘Calculate
Coeff.’ keypress, the system will switch to the
wide Doppler beam, and the keymap associated
with the measurement stage will appear.
(5) After delineation of the laminar flow portion of the
measurement beam Doppler spectrum, and select-
ing a cardiac cycle of interest, upon pressing the
‘Analyze Cycle’ key, the software computes the
PVI to yield estimates of instantaneous flow rates,
1
Transmit pulse repetition frequency (PRF) sufficiently high
to eliminate ambiguity of blood flow direction and velocities
(aliasing) when measuring velocities larger than 300 cm/s by
exchanging for an ambiguity in the depth because of more than
one sample volume.
Int J Cardiovasc Imaging
123
and finally the flow rates are integrated over the
selected time interval of interest to produce a
measure of total RSV within a cardiac cycle.
The RSV measurement, along with the average
CSA of flow, are then displayed onscreen to the
user.
(2) Matrix-array transducer
For initial application of matrix-array transducer
design we used an existing 3.5 MHz probe (Philips
Medical Systems, Andover, MA, USA) with multiple
separate bars of 64 piezoelectric elements each in the
elevation dimension to achieve the following require-
ments: (1) generation of a narrow reference beam and
a broad measurement beam, (2) electronic beamwidth
control, and (3) elimination of transmit power
discrepancy between the narrow and the broad beam.
In conventional phased-array transducers the elec-
tronic steering and focusing beamformer component
focuses only in the lateral dimension the receive
ultrasound beam at the desired spatial location within
the body. Because the matrix-array transducer con-
sists of separate elements in both, lateral and
elevation dimensions that can be electronically turned
on and off (Fig. 4) it affords control over both, the
lateral and elevation apertures.
To meet the requirements of the PVI method the
following combination of settings for the matrix
probe was used: For the broad measurement beam,
turning off the outer element bars (Y-groups) and
leaving only the center twenty four elements active
across the lateral dimension in both the transmit and
receive directions effectively reduced the transducer
aperture to 37.5% in the lateral dimension and 50% in
the elevation dimension (Fig. 4), thus creating a
broader beam in the far field as a result of a wide
overall roundtrip beam response. In the case of the
narrow Doppler calibration beam, instead of using the
full aperture on both transmit and receive response,
we used a method by which the transducer produced
anarrow overall roundtrip beam response by trans-
mitting a broad beam while generating a narrow
beam on receive (Figs. 4and 5). Note, that only
where the transmit and receive beams overlap will
there be an appreciable acoustic response detected by
the probe. As a result of this approach transmit power
remained constant when electronically switching
from the calibration to the measurement beam,
thereby eliminating the need for calculations required
to correct for differing transmit powers that would
have been extremely complex and prone to error.
Implementation of the broad transmit beam for
calibration was achieved identically to the measure-
ment beam, whereas the narrow receive beam for
Fig. 4 Diagrams of the
different transducer
apertures of the matrix-
array probe; highlighted
portions represent the active
transducer elements in each
state. Above, The reduced
aperture is used to produce
the broad transmit beam in
both calibration and receive
stages. Below, On receive,
the full aperture is used to
generate the narrow receive
beam during calibration,
while the reduced aperture
forms the wide receive
beam during the
measurement stage
Int J Cardiovasc Imaging
123
calibration is implemented by leaving active the full
aperture in both the lateral and elevation dimensions
(Figs. 4and 5). Thus, a correction factor (CF) was
required to correct only the power from the mea-
surement beam for the decrease in receive
sensitivities resulting from aperture reduction. CF
was obtained using a computer modeling program
simulating transducer aperture reduction to predict
the energy received by the narrow and wide beams at
varying depths; CF determined to be 23.734 at a
depth of 10 cm due to energy received by narrow
beam of 2.324E+05 J/cm
2
and energy received by
wide beam of 9.791E+03 J/cm
2
.
In addition, for both beams, a flat (i.e., uniform)
apodization profile across the lateral elements of
the transducer was applied; apodization referring
to the process of applying an independent multipli-
cative weight to the pressure field response at each
piezoelectric element in the transducer to affect the
spatial response of the transducer array.
Validation
Testing and validation of the system was divided into
three distinct rounds of experimentation: (1) initial
signal generator trials for verification of the imple-
mented Doppler power analysis software, (2) flow
phantom trials providing known values of flow
velocity, flow orifice size, and total flow volume,
against which to compare Doppler power measure-
ments, and (3) final in vivo trials to assess the system
performance in the clinical environment.
Signal generator trial
A signal generator was set to produce a sinusoid with
an approximate frequency of 1.9 MHz and variable
amplitudes adjustable in 1 volt increments. The
signal was applied directly to the transmit and receive
channels of the ultrasound system front end. The
experiment performed with the signal generator setup
tested the entire system integration of all the various
calculations and corrections contained. Input from the
signal generator produced a single, constant tone on
the ultrasound system’s display (Fig. 6). By tuning
the frequency of the signal generator output so that
the Doppler waveform was situated at a labeled
velocity on the display, a calculation of a simulated
flow volume was possible. For this, the combination
of Eqs. 9,13, and 14 yielded a complete formula for
calculating total flow volume:
Qv¼CF CSAref
Pref
Z
TZ
vel
vPmeas vðÞdvdt ð15Þ
Thus, if an identical input signal was used in both
the calibration and measurement stages, then the total
backscattered powers received by the narrow and
wide beams should be related by the following
equation:
Pref ¼CF
Pmeas ð16Þ
Realizing that ‘‘flow velocity’’ and ‘‘backscattered
power’’ are constant over time due to the signal
generator acting as the input source, and taking note
Fig. 5 Basic principle of overall roundtrip beam response.
Only where the transmit and receive beams overlap will there
be an appreciable acoustic response detected by the probe.
Thus, transmitting a broad beam and receiving a narrow beam
produced the narrow overall roundtrip beam required for
calibration (above), whereas a broad beam in both the transmit
and receive directions produced a broad overall roundtrip beam
required for measurement (below)
Int J Cardiovasc Imaging
123
of Eq. 16, Eq. 15 accordingly simplified to the
following:
Qv¼CF CSAref
CF
Pmeas
vZ
TZ
vel
Pmeas vðÞdvdt
¼vCSAref
Pmeas
Z
T
Pmeasdt ð17Þ
Finally, because all data values resided within a
discrete time domain, the transformation below
needed to be applied:
Qv¼vCSAref
Pmeas
Z
T
Pmeasdt
vCSAref
Pmeas
X
T
Pmeas DtðÞ
¼vCSAref
Pmeas
DtX
T
Pmeas
¼vCSAref
Pmeas
DtT
Pmeas
¼vCSAref DtT
ð18Þ
All the remaining variables in Eq. 18 were known
values. Absolute velocity, v, could be read off of
Fig. 6as 120 cm/s. Cross-sectional area of the narrow
beam, CSA
ref
, at a depth of 10.1 cm was 0.031465p
cm
2
. Since the sweep speed was set at 50 mm/s in the
experiment, Dt, the amount of time represented by a
single column of spectral data on the Doppler display,
was known to be 0.01 s. Lastly, Twas the number of
discrete time units over which the simulated flow
extended, and was set to be 30 in the current
experiment. Substituting all known values into Eq.
18 yielded a simulated absolute flow volume of
3.559 cc. And as desired, the following measurement
made with the Doppler power software package, as
shown in Fig. 6, almost perfectly matched the theo-
retical result above (3.587 cc vs. 3.559 cc).
Flow phantom trials
The second round of validation was performed using
a previously described flow phantom (Fig. 7) with
known physical characteristics, against which to test
the accuracy of the computed measurements [16].
Flow of blood-mimicking fluid was measured by the
PVI method and directly compared to known steady
flow rates of 10, 20, 30, 40 and 50 ml/s from a motor-
driven piston pump that minimized cavitation (mod-
ified Mark IV Power-injector, Medrad, Indianola, PA)
as well as parabolic pulses of 10, 20, 30, 40 and 50 ml
from manual syringe injections, all passing through
circular orifices of 0.24, 0.38, 0.5, 0.57, 0.64 and
0.78 cm
2
(diameters of 0.55–1.0 cm, corresponding to
clinical lesions ranging from ‘‘mild to severe’’) [18].
Fig. 6 Doppler waveform
produced by the signal
generator, simulating
laminar flow at a constant
absolute velocity of
120 cm/s and at a depth of
10.1 cm
Int J Cardiovasc Imaging
123
Practical application of the system provided suc-
cessful acquisition of laminar VC Doppler velocity
spectra in both software components, the calibration
and the measurement stage, with reliable velocity
border detection in most flow conditions. Automated
PVI analysis in the calibration stage and in the
measurement stage was performed in less than 10 s
each. In Doppler signals where automated velocity
border detection failed an addition of 15 s was
required for manual border tracings in each stage.
Steady flow rates and pulsatile stroke volumes by the
PVI method correlated (r=0.98, y=0.97x +0.9
ml/s; r=0.99, y=0.98x +0.8 ml) and agreed
well (SEE =2.4 ml/s; SEE =2.2 ml) with actual
values for orifices up to 0.85 cm diameter or 0.57 cm
2
CSA. Flow rates and stroke volumes passing through
circular orifices of 0.9 cm diameter and larger were
underestimated because orifices of this size, which are
clinically extreme, were incompletely encompassed
with current beam width. Orifice areas were system-
atically underestimated by PVI according to a factor
of 0.74 as it was expected for effective versus
anatomic EROA due to a flow contraction coefficient
of 0.65–0.85 as described in prior literature [24].
Small circular orifices with 0.5 cm diameter (equals
EROA 0.2 cm
2
) and less, with an effective diameter
of 0.44 cm (equals EROA 0.15 cm
2
; 0.2 cm
2
90.74
=0.15 cm
2
) and less only allowed incom-
plete inclusion of the currently available narrow
calibration beam within the VC flow as a results of
estimated beam dimensions of 0.3 cm (lateral
dimension) times 0.4 cm (elevation dimension) at
10 cm depth.
In vivo trials
Clinical validation involved 48 patients (age
60.7 ±16.4 years; 29 male, 18 female) with MR of
different etiologies, studied from a transthoracic
apical approach with at least fair to good 2D and
color Doppler image quality. Calculated RSV was
compared with MRI obtained within 1 h of the
Doppler study with a 1.5 Tesla system (Magnetom
Sonata, Siemens Medical Systems, Erlangen, Ger-
many). Mitral RSV was obtained as mitral inflow
minus aortic outflow from phase-contrast velocity
maps [25,26]. Phase-contrast cine acquisitions were
obtained in planes aligned with the mitral annulus
and orthogonal to the mid-ascending aorta. An ECG-
triggered free-breathing through-plane phase-contrast
sequence (Repetition time 25 ms; Echo time 4.8 ms;
flip angle 15°; matrix 129 9256; four averages) was
used.
Of all patients 23 had functional MR, 11 had
degenerative MR without mitral valve prolapse
(MVP), and 14 had MVP. 18 patients had an
eccentric regurgitant jet (MVP: 13, degenerative
MR: 3, functional MR: 2). RSV by PVI correlated
Fig. 7 Diagram and
photograph of the flow
phantom used for
experimental validation
studies. The flow phantom
consisted of a variable
diameter orifice located
10 cm from the ultrasound
transducer acoustic window
(distance in vivo of a mitral
valve from the transducer
placed on the chest surface),
with flow passing from a
cylindrical Plexiglas
chamber (5.7 cm diameter;
mimicking the left
ventricle) to an unconfined
receiving chamber
(mimicking the left atrium)
Int J Cardiovasc Imaging
123
(r=0.93, y=0.88x +0.33 ml) and agreed well
(mean error =0.25 ±0.33 ml) with MRI reference
values up to 40 ml with significant underestimation
only at higher RSVs with an EROA size [0.65 cm
2
,
which was consistent with the underestimation due to
currently limited beam width (Fig. 8). Agreement of
RSV by PVI and MRI was also good in patients with
eccentric MR jets with no significant difference
between central or eccentric MR (mean difference
0.32 ±3.8 ml vs. 0.39 ±4.0 ml, NS by ttest). That
is because the PVI method is relatively immune to
variations in the Doppler beam-to-flow angle has the
decrease in measured velocity due to cos his
canceled by a reciprocal increase in CSA relative to
the beam, as demonstrated before in a group of
patients including eccentric regurgitant jets and
confirmed by the present study results [18].
Discussion
To accomplish clinical application of the recently
introduced concept of backscattered Doppler power
times velocity at the VC for direct measurement of
regurgitant flow in valvular heart disease, we
described the practical implementation of the PVI
method on a clinically used matrix-array ultrasound
imaging system including implementation of pro-
found software modifications for automated Doppler
flow analysis. Implementation of the PVI method on
a system using matrix-array transducer technology
was particularly realized by achieving wide overall
roundtrip beam response of the Doppler measure-
ment beam by transmitting a broad beam and forming
the same broad beam on receive, whereas narrow
overall roundtrip beam response of the Doppler
calibration beam was achieved by transmitting the
same broad beam while generating a narrow beam on
receive.
Experimental in vitro and in vivo validation
studies of a wide range of clinically relevant flow
conditions have recently demonstrated a high degree
of accuracy between flow measurements by the PVI
method and true measures of flow or reference flow
values by MRI [18]. Though the present study was
focusing on testing and validation of an integrated
and automated PVI method on a matrix-array ultra-
sound system, similar accuracy of PVI measurements
was found in an even larger group of patient with MR
of different etiologies. One advantage of the PVI
method contributing to the high accuracy of flow
measurements in a broad spectrum of regurgitant
lesion sizes and shapes as well as central and
eccentric jet formation is the uniform beam response
of a nearly circular, broad sample volume in both,
lateral and elevational dimension as previously
demonstrated in computer models, indicating uniform
spatial resolution of the matrix-array transducer in
both dimensions [16]. This uniform sonification has
been demonstrated to provide accurate flow quanti-
fication independent of the shape and size of the
encompassed flow CSA, this being particularly
important in non-circular regurgitant orifices [16].
As a second advantage of the PVI method, because
the cos hdecrease in measured velocity is canceled
by a reciprocal increase in CSA, accuracy has been
demonstrated to be persistent also for eccentric jets,
which have been difficult to evaluate based on jet
Fig. 8 Patient study results for calculated RSV values versus
MRI. Values affected by systematic underestimation of higher
RSVs (white circles) due to limited beam width were excluded
from linear regression analysis
Int J Cardiovasc Imaging
123
area or using continuous-wave Doppler in previous
studies [18].
After acquisition of a characteristic, narrow Dopp-
ler velocity spectrum at the VC, automated Doppler
signal analysis in the calibration stage and in the
measurement stage was performed in less than 10 s
each. This Doppler power flow analysis has the major
advantage of providing automated noninvasive flow
measurement in valvular heart disease by means of a
self-operating software-controlled Doppler signal
analysis and flow calculation that is completely user
independent for the first time. Thus, the presented
automated Doppler power flow analysis overcomes
practical limitations of existing ultrasound methods,
either requiring multiple-step 2D and Doppler mea-
surements apart form the lesion itself or color-
Doppler measurements of the proximal flow conver-
gence [8,10–12,27,28]; all those manual onscreen
measurements being highly dependent on user skills
and expertise and, therefore, being a source of
significant variability in clinical routine [29].
Current limitations and future perspectives
Automated flow analysis in the stage of development
as presented was mainly limited by the separation of
the calibration and measurement stage, this separa-
tion causing potential inaccuracy by differences in
attenuation and flow condition between the Doppler
signals from separate cardiac cycles. Through the use
of parallel signal processing we foresee the following
future application: On transmit the transducer gener-
ates a broad and relatively uniform beam that sonifies
the entire area of flow. On receive, the two required
beams, the narrow reference and the broad measure-
ment beam, are generated simultaneously by the
connection of the transducer elements to two inde-
pendent digital beamforming processors. Thus, the
Doppler signal information could be acquired and
calibrated in a single cardiac phase, thereby elimi-
nating the uncertainty about the consistency of two
separate Doppler signals. By means of simultaneous
generation of a narrow and a broad Doppler beam, yet
the current limitation of poor 2D image resolution in
the measurement stage due to the unfocused, broad
beam, could be overcome by applying the narrow
beam for both calibration and 2D imaging (for
navigation purposes), while using the broad beam
for power measurement. Ideally, navigation of the
sample volume into the VC of a high-velocity flow jet
can be fully automated by an algorithm steering the
narrow beam sample volume and searching for the
narrowest, high-velocity signal with the largest PVI,
which unambiguously defines the position within the
VC, thereby obviating the need for 2D imaging [30].
Although the presented matrix-array technique
provides a maximal beam width of nearly 1 cm
diameter, which is enough to cover most clinical
cases of MR, ideally the beam should cover areas up
to 2 cm diameter in order to reliably encompass also
EROAs that are elongated along the non-circular or
slit-like commissural line between the two mitral
leaflets as typical in functional mitral regurgitation. In
principle, a wider and more uniform beam should be
achieved with recently introduced 2D matrix-array
probes, consisting of a transducer array with a total of
3,000 active elements providing a sufficiently high
number of elements in both dimensions, lateral and
elevation dimension [19]. Ultimately, a beam capable
of encompassing the entire mitral and aortic valves
would be desired for numerous other applications of
the PVI method, including but not limited to the
following: aortic and tricuspid regurgitation, mitral
and aortic stenosis, shunt lesions, mitral inflow, and
aortic outflow.
Ultimately, we envision the following practical
application of automated Doppler power flow anal-
ysis in routine clinical practice: A system with a 2D
matrix-array probe automatically navigates the nar-
row beam sample volume into the VC where online
Doppler signal analysis detects the narrowest, high-
velocity signal with the largest PVI. As a secondary
effect, navigation of the narrow beam within the flow
jet ensures optimal sonification of the flow CSA by
the broad beam. Finally, simultaneous acquisition of
the narrow reference beam and the broad measure-
ment beam enables instantaneous power calibration;
thus, providing continuous online calculation and
onscreen display of RSV and dynamic changes of
flow CSA within each cardiac cycle.
Conclusions
The new method of power–velocity integration for
direct, noninvasive quantification of flow in valvular
heart disease can be fully implemented on board a
self-contained, routinely used diagnostic ultrasound
Int J Cardiovasc Imaging
123
system. Through this, individual steps of Doppler
spectral analysis can be automated providing rapid,
user-independent quantification of regurgitant flow
volume and EROA in routine clinical practice for the
first time. Such automated Doppler power flow
analysis of valvular regurgitant flow directly, nonin-
vasively, and user independent overcomes the
practical limitations of current techniques that are
due to technically demanding and time-consuming
manual onscreen measurements. This automated
Doppler power flow analysis can be readily imple-
mented in existing digital ultrasound systems using
new generation 2D matrix-array transducer design.
Acknowledgements T. Buck was supported by grants Bu1097/
2-1 and Bu 1097/2-2 fromthe Deutsche Forschungsgemeinschaft,
Bonn, Germany. S.M. Hwang was a student of Electrical
Engineering and Computer Science at the Massachusetts
Institute of Technology in the years 1999 till 2002. The work
was supported in part by NIH grants R01HL38176, HL53702, and
K24 HL67434 of the National Institutes of Health, Bethesda,
Maryland to R.A. Levine.
References
1. Hatle L, Angelsen BAJ (1985) Doppler ultrasound in car-
diology: physical principles and clinical applications.
Lea&Febiger, Philadelphia
2. Singh JP, Evans JC, Levy D, Larson MG, Freed LA, Fuller
DL, Lehmann B, Benjamin EJ (1999) Prevalence and
clinical determinants of mitral, tricuspid, and aortic
regurgitation (The Framingham Heart Study). Am J Car-
diol 83:897–902
3. Iung B, Baron G, Butchart EG, Delahaye F, Gohlke-Bar-
wolf C, Levang OW, Tornos P, Vanoverschelde JL,
Vermeer F, Boersma E, Ravaud P, Vahanian A (2003) A
prospective survey of patients with valvular heart disease
in Europe: the Euro heart survey on valvular heart disease.
Eur Heart J 24:1231–1243
4. Eckberg DL, Gault JH, Bouchard RL, Karliner JS, Ross J Jr
(1973) Mechanics of left ventricular contraction in chronic
severe mitral regurgitation. Circulation 47:1252–1259
5. Zoghbi WA, Enriquez-Sarano M, Foster E, Grayburn PA,
Kraft CD, Levine RA, Nihoyannopoulos P, Otto CM,
Quinones MA, Rakowski H, Stewart WJ, Waggoner A,
Weissman NJ (2003) Recommendations for evaluation of
the severity of native valvular regurgitation with two-
dimensional and Doppler echocardiography. J Am Soc
Echocardiogr 16:777–802
6. Quinones MA (1998) Management of mitral regurgitation.
Optimal timing for surgery. Cardiol Clin 16:421–435
7. Enriquez-Sarano M, Avierinos JF, Messika-Zeitoun D,
Detaint D, Capps M, Nkomo V, Scott C, Schaff HV, Tajik
AJ (2005) Quantitative determinants of the outcome
of asymptomatic mitral regurgitation. N Engl J Med
352:875–883
8. Recusani F, Bargiggia GS, Yoganathan AP, Raisaro A,
Valdez-Cruz L, Sung HW, Bertucci C, Gallati M, Moises
V, Simpson IA, Tronconi L, Sahn DJ (1991) A new
method for quantification of regurgitant flow rate using
color flow imaging of the flow convergence region proxi-
mal to a discrete orifice: an vitro study. Circulation
83:594–604
9. Bolger AF, Eigler NL, Maurer G (1988) Quantifying val-
vular regurgitation: the limitations and inherent
assumptions of Doppler techniques. Circulation 87:1316–
1318
10. Utsunomiya T, Ogawa T, Doshi R, Patel D, Quan M,
Henry WL, Gardin JM (1991) Doppler color flow ‘‘prox-
imal isovelocity surface area’’: method for estimating
volume flow rate: effects of orifice shape and machine
factors. J Am Coll Cardiol 17:1103–1111
11. Enriquez-Sarano M, Bailey KR, Seward JB, Tajik AJ,
Krohn MJ, Mays JM (1993) Quantitative Doppler assess-
ment of valvular regurgitation. Circulation 87:841–848
12. Blumlein S, Bouchard A, Schiller NB, Dae M, Byrd BF III,
Ports T, Botvinick EH (1986) Quantitation of mitral
regurgitation by Doppler echocardiography. Circulation
74:306–314
13. Grayburn PA, Peshock RM (1996) Noninvasive quantifi-
cation of valvular regurgitation. Circulation 94:119–121
14. Hottinger CF, Meindl JD (1979) Blood flow measurement
using the attenuation compensated volume flowmeter.
Ultrason Imaging 1:1–15
15. Yoganathan AP, Cape EG, Sung HW, Williams FP, Jimoh
A (1988) Review of hydrodynamic principles for the car-
diologist: applications to the study of blood flow and jets
by imaging techniques. J Am Coll Cardiol 12:1344–1353
16. Buck T, Mucci RA, Guerrero JL, Holmvang G, Hands-
chumacher MD, Levine RA (2000) Flow quantification in
valvular heart disease based on the integral of backscat-
tered acoustic power using Doppler ultrasound. Proc IEEE
88:307–330
17. Buck T, Mucci RA, Guerrero JL, Holmvang G, Hands-
chumacher MD, Levine RA (2000) The power-velocity
integral at the vena contracta – a new method for direct
quantification of regurgitant volume flow. Circulation
102:1053–1061
18. Buck T, Plicht B, Hunold P, Mucci RA, Erbel R, Levine
RA (2005) Broad-beam spectral Doppler sonification of the
vena contracta using matrix-array technology – A new
solution for semi-automated quantification of mitral
regurgitant flow volume and orifice area. J Am Coll Card
45:770–779
19. Sugeng L, Weinert L, Thiele K, Lang RM (2003) Real-
time three-dimensional echocardiography using a novel
matrix array transducer. Echocardiography 20:623–635
20. Angelsen BAJ (1980) A theoretical study of the scattering
of ultrasound from blood. IEEE Trans Biomed Eng 27:61–
67
21. Shung KK, Cloutier G, Lim CC (1992) The effects of
hematocrit, shear rate, and turbulence on ultrasonic
Doppler spectrum from blood. IEEE Trans Biomed Eng
39:462–469
Int J Cardiovasc Imaging
123
22. Brody WR, Meindl JD (1974) Theoretical analysis of the
CW Doppler ultrasound flowmeter. IEEE Trans Biomed
Eng 21:183–192
23. Looyenga DS, Liebson PR, Bone RC, Balk RA, Messer JV
(1989) Determination of cardiac output in critically ill
patients by dual beam Doppler echocardiography. J Am
Coll Cardiol 13:340–347
24. Daugherty RL, Franzini JB (1977) Fluid mechanics with
engineering applications. McGraw-Hill, New York, NY
25. Fujita N, Chazouilleres AF, Hartiala JJ, O’Sullivan M,
Heidenreich P, Kaplan JD, Sakuma H, Foster E, Caputo
GR, Higgins CB (1994) Quantification of mitral regurgi-
tation by velocity-encoded cine nuclear magnetic
resonance imaging. J Am Coll Cardiol 23:951–958
26. Hundley WG, Li HF, Willard JE, Landau C, Lange RA,
Meshack BM, Hillis LD, Peshock RM (1995) Magnetic
resonance imaging assessment of the severity of mitral
regurgitation: comparison with invasive techniques. Cir-
culation 92:1151–1158
27. Rokey R, Sterling LL, Zoghbi WA, Sartori MP, Limacher
MC, Kuo LC, Quinones MA (1986) Determination of
regurgitant fraction in isolated mitral or aortic regurgitation
by pulsed Doppler two-dimensional echocardiography. J
Am Coll Cardiol 7:1273–1278
28. Enriquez-Sarano M, Miller FA Jr, Hayes SN, Bailey KR,
Tajik AJ, Seward JB (1995) Effective mitral regurgitant
orifice area: clinical use and pitfalls of the proximal isove-
locity surface area method. J Am Coll Cardiol 25:703–709
29. Flachskampf FA, Frieske R, Engelhard B, Grenner H,
Frielingsdorf J, Beck F, Reineke T, Thomas JD, Hanrath P
(1998) Comparison of transoesophageal Doppler methods
with angiography for evaluation of the severity of mitral
regurgitation. J Am Soc Echocardiogr 11:882–892
30. Buck T (2007) Method and device for ultrasound mea-
surement of blood flow. Tracking of blood flow
measurement. Patent DE 103.12.883 B4. Date issued
22.02.2007
Int J Cardiovasc Imaging
123
